Core Insights - The core viewpoint is that 2026 will mark the second half of embodied intelligence, focusing on application maturity on both supply and demand sides [1][34]. Group 1: Development and Challenges - The embodied intelligence industry is currently perceived to be in a "technical bottleneck" phase, with a significant need for high-quality data and a correct ontology to drive progress [3][4][10]. - The lack of high-quality data is attributed to the absence of a standard ontology, which is essential for effective data collection and model training [4][11]. - The current focus should be on achieving object and action generalization, as scene and ontology generalization are less critical at this stage [17][27]. Group 2: Product and Model Structure - The ideal product structure for embodied intelligence is a combination of "hardware + pre-trained models + post-training tools," which allows for effective task execution in specific environments [6][9]. - The training process involves a two-phase approach: pre-training to understand basic interactions with the physical world and post-training for specific tasks [21][23]. - The model architecture includes a "slow thinking" component for logical reasoning and a "fast execution" component for real-time actions, which is crucial for operational efficiency [22][19]. Group 3: Market Dynamics and Future Outlook - The market is returning to a rational state, with companies exploring practical applications of embodied intelligence rather than unrealistic expectations of immediate widespread adoption [33][34]. - By 2026, the supply side will see matured robot bodies and initial generalization capabilities, while the demand side will have clearer application scenarios [32][34]. - The growth of the developer community is essential for the prosperity of the embodied intelligence market, as they will create diverse applications that drive value [28][29].